On rainbow matchings for hypergraphs

TL;DR

This paper proves a conjecture on rainbow matchings in hypergraphs for large enough parameters and establishes tight bounds on the product of set sizes avoiding rainbow matchings.

Contribution

It confirms the conjecture for n ≥ 3(k-1)(t-1) and determines the maximum product of sizes of hypergraph families avoiding rainbow matchings.

Findings

Conjecture holds when n ≥ 3(k-1)(t-1).

Established tight bounds on the product of set sizes.

Provided conditions under which rainbow matchings do not exist.

Abstract

For any posotive integer $m$, let $[m]:=\{1,\ldots,m\}$. Let $n,k,t$ be positive integers. Aharoni and Howard conjectured that if, for $i\in [t]$, $\mathcal{F}_i\subset[n]^k:= \{(a_1,\ldots,a_k): a_j\in [n] \mbox{ for } j\in [k]\}$ and $|\mathcal{F}_i|>(t-1)n^{k-1}$, then there exist $M\subseteq [n]^k$ such that $|M|=t$ and $|M\cap \mathcal{F}_i|=1$ for $i\in [t]$ We show that this conjecture holds when $n\geq 3(k-1)(t-1)$. Let $n, t, k_1\ge k_2\geq \ldots\geq k_t $ be positive integers. Huang, Loh and Sudakov asked for the maximum $\Pi_{i=1}^t |{\cal R}_i|$ over all ${\cal R}=\{{\cal R}_1, \ldots ,{\cal R}_t\}$ such that each ${\cal R}_i$ is a collection of $k_i$-subsets of $[n]$ for which there does not exist a collection $M$ of subsets of $[n]$ such that $|M|=t$ and $|M\cap \mathcal{R}_i|=1$ for $i\in [t]$ %and ${\cal R}$ does not admit a rainbow matching. We show that for sufficiently large $n$ with $\sum_{i=1}^t k_i\leq n(1-(4k\ln n/n)^{1/k}) $, $\prod_{i=1}^t |\mathcal{R}_i|\leq {n-1\choose k_1-1}{n-1\choose k_2-1}\prod_{i=3}^{t}{n\choose k_i}$. This bound is tight.